Common Cause Failure and Model Construction.

The model we have been using so far is a simple tree shape, there are no (undirected) loops in the tree. This, is by no means, a limitation of Graphical-Belief. Graphical-Belief's model compilation procedure allows the user to build models with common-causes: single variables which serve as an input to more than one variable. Although adding loops adds to the computational complexity, Graphical-Belief can still handle the resulting models. This example also allows us to explore some of the model construction features of Graphical-Belief.

For the purposes of the example, suppose we learn that the Motor Operated Valves have a higher failure rate when exposed to live steam. As presumable live steam would fill the room containing both MOV-25-A and MOV-25-B, both would be effected. Because the common cause could cause a failure in both redundant branches in the system, the presence of such a common cause failure could dramatically increase the system failure rate and we need to carefully study it. For the purposes of illustration, we will assume that the probability live steam is present during an accident is .01 and that the MOVs are ten times more likely to fail if live steam is present.

Model Construction

To add the common cause failure to the model we must:

1. Add a new variable to represent the presence of live steam.

2. Create a rule which describes the distribution of the live steam variable.

3. Create a new class of rules which describes the distribution of MOV failures conditioned on the presence of live steam.

4. Compile the model (to get rid of the loops) and analyze the results.

Knowledge Engineering

This example has given us a chance to explore the model construction features of Graphical-Belief. Here we see some of its unique knowledge engineering features:
  1. We can draw from a library of previously created variables and rules, drawing on work and knowledge stored in previous models.
  2. We can create new classes of variables and rules so that we can re-use the same knowledge in many places and we can keep track of where we used that knowledge.
  3. We can set the signature of rules so that they are only used in appropriate contexts.
  4. We can save graph fragments as rules so that we can re-use portions of our model [we are currently working on this feature.]
These features provide a powerful mechanism for sharing expertise among a team of analysts. In many cases only one or two analysts need understand the full complexity of each rule. Other analysts can draw on their previous work. This makes Graphical-Belief a very powerful environment for exploring graphical models.

Another part of Graphical-Belief's power is not apparent from any of the model we have so far explored: Graphical-Belief can use both probabilities and belief functions as the primary representation of uncertainty. The next example explore this flexibility.

Valuations. Graphical-Belief can use many representations of uncertainty, this example explores a few.

Return to the main example page.

Back to overview of Graphical-Belief.

View a list of Graphical-Belief in publications and downloadable technical reports.

The Graphical-Belief user interface is implemented in Garnet.

Get more information about obtaining Graphical-Belief (and why it is not generally available).

get the home page for Russell Almond , author of Graphical-Belief.

Click here to get to the home page for Insightful (the company that StatSci has eventually evolved into).

Russell Almond, <lastname> (at) acm.org
Last modified: Fri Aug 16 17:37:55 1996